Density Operator and Mixed States

Every state in this track so far has been a ket — a maximal description, everything quantum mechanics permits you to know. Laboratories are rarely so lucky. An oven spits out silver atoms with random spins; a laser's output flickers among nearby modes; your qubit talked to a stray field you didn't monitor. In each case the honest description is a statistical ensemble: with probability $p_1$ the system is in $|\psi_1\rangle$ , with $p_2$ in $|\psi_2\rangle$ — classical ignorance stacked on top of quantum amplitudes. This article builds the object that handles both layers in one package: the density operator $\rho = \sum_i p_i|\psi_i\rangle\langle\psi_i|$ , with every prediction delivered by one formula, $\langle\hat A\rangle = \mathrm{Tr}(\rho\hat A)$ . The centerpiece is a distinction your hands must learn: a mixture is not a superposition. Fifty percent $|{+z}\rangle$ plus fifty percent $|{-z}\rangle$ is not the state $|{+x}\rangle = (|{+z}\rangle + |{-z}\rangle)/\sqrt2$ — the superposition holds a phase relationship the mixture lacks, and one measurement axis exposes them: $\langle S_x\rangle = \hbar/2$ for the pure state, $0$ for the mixture. The density matrix wears the difference on its face (off-diagonal coherences present or absent), the purity scalar $\mathrm{Tr}\rho^2$ grades it ( $1$ versus $\tfrac12$ ), and for qubits the whole state space becomes a picture — the Bloch ball, pure states on the skin, mixtures in the flesh, the fully ignorant state at the center. Two more lessons complete the kit. Ensembles are non-unique: mixing $z$ -spins fifty-fifty and mixing $x$ -spins fifty-fifty produce the same ρ — physically indistinguishable forever, a deep fact about what quantum states are. And ρ is forced on you by entanglement: half of a Bell pair, computed by partial trace, is maximally mixed even though the whole is pure — the one-card preview of the chapter this track ends with, where ρ graduates from bookkeeping to protagonist.

Key takeaways

$\rho = \sum_i p_i|\psi_i\rangle\langle\psi_i|$ : classical mixing weights $p_i$ over quantum states; every observable via $\langle\hat A\rangle = \mathrm{Tr}(\rho\hat A)$ .
Three defining properties (derived): Hermitian, $\mathrm{Tr}\rho = 1$ , positive semidefinite — anything satisfying them is a legal state.
Purity test: pure $\iff \rho = |\psi\rangle\langle\psi| \iff \rho^2 = \rho \iff \mathrm{Tr}\rho^2 = 1$ ; the minimum $\mathrm{Tr}\rho^2 = 1/d$ is the maximally mixed $\mathbb 1/d$ .
Mixture ≠ superposition: 50/50 $z$ -mixture gives $\rho = \mathbb 1/2$ with $\langle S_x\rangle = 0$ ; the pure $|{+x}\rangle$ gives $\langle S_x\rangle = \hbar/2$ . Coherences (off-diagonals) are the receipts.
Bloch ball: $\rho = \tfrac12(\mathbb 1 + \mathbf r\cdot\boldsymbol\sigma)$ , eigenvalues $(1 \pm |\mathbf r|)/2$ , legal iff $|\mathbf r| \le 1$ ; purity $= (1 + r^2)/2$ ; surface = pure, center = unpolarized.
Non-uniqueness: different ensembles can share one ρ (z-mixture = x-mixture = $\mathbb 1/2$ ) — and same ρ means no experiment whatsoever distinguishes the preparations.
Partial trace preview: half a Bell pair is exactly $\mathbb 1/2$ — pure whole, maximally mixed part; the entanglement chapter owns the theory.
Key idea: one positive, unit-trace, Hermitian operator carries both quantum amplitudes and classical ignorance; everything measurable is $\mathrm{Tr}(\rho\hat A)$ .

The core insight

Pure and mixed states are unified by a single object ρ: a positive, unit-trace, Hermitian operator. Everything measurable is $\mathrm{Tr}(\rho\hat A)$ , purity is graded by $\mathrm{Tr}\rho^2$ , and mixtures — unlike superpositions — carry no coherences between their branches.

Where this lives today

The density matrix is the native language of every quantum technology's data sheet. State tomography — the workhorse verification of quantum processors — is nothing but measuring enough expectation values to reconstruct ρ, then quoting its purity and its fidelity to the target; the Bloch/Stokes vector of Example 2 is the one-qubit case, and IBM, Google, and IonQ report multi-qubit generalizations daily. Decoherence budgets are coherence bookkeeping: $T_2$ times measure how fast off-diagonal elements die (the warning box's receipts, decaying in real time), and error models are maps on ρ — the Lindblad machinery seeded by the von Neumann equation this article deferred by name. NMR and MRI read the Bloch vector literally: the magnetization of your tissue is $\mathbf r$ scaled by spin density, precessing and relaxing toward the ball's interior — Bloch's 1946 equations run on hospital hardware. And at the conceptual frontier, the purity of reduced states quantifies entanglement in many-body simulations and quantum gravity toy models alike — the partial-trace preview grown into a research industry this track's final chapter opens.

The physical picture

Picture the difference as two ways of not knowing where a spin points. In a superposition $|{+x}\rangle$ , the spin genuinely holds both $z$ -answers in phase-locked suspension — and the phase is physical: rotate your apparatus to the $x$ -axis and you get $+\hbar/2$ every single time. In the mixture, each atom left the oven already committed to $+z$ or $-z$ ; you merely weren't told which. No axis gives certainty: the ignorance is classical, sitting on top of the quantum states rather than inside them. The density operator is the ledger that records exactly this difference — and nothing else: two preparations with the same ledger are the same state, full stop.

Figure

Mixture versus superposition — same z-statistics, different worlds

Where do mixtures come from? Three factories, in ascending depth. Ignorant preparation: ovens, thermal sources, noisy hardware — probabilities from not controlling the source. Unread measurement: the collapse article's $\rho' = \sum_n\hat P_n\rho\hat P_n$ — probabilities from not reading the record. And — the deep one — entanglement: look at part of a composite pure state and the part is mixed with no ignorance anywhere in the universe; the information sits in correlations you cannot see locally. The first two are conveniences; the third, previewed in the last derivation below, is why ρ is a necessity rather than a luxury — and why this article's register-twin at the track's end (The Density Operator and Mixed States, in the entanglement chapter) re-derives everything here with entanglement as the starting point rather than the finale.

The object and its credentials

Let a source produce $|\psi_i\rangle$ with probability $p_i$ (the $|\psi_i\rangle$ normalized, not necessarily orthogonal; $\sum p_i = 1$ ). Define

\rho \;=\; \sum_i p_i\,|\psi_i\rangle\langle\psi_i| \qquad (1)

where: each term is the projector onto one possible identity (the outer-product objects of the projector article), weighted by its classical odds. A pure state is the one-term case $\rho = |\psi\rangle\langle\psi|$ . The definition earns its keep through one theorem:

Derivation AEverything measurable is a trace — and the three credentials of a legal state

We derive $\langle\hat A\rangle = \mathrm{Tr}(\rho\hat A)$ from the ensemble definition and establish the three properties — Hermiticity, unit trace, positivity — that characterize density operators without reference to any ensemble. The strategy: average the pure-state Born statistics over the classical weights, recognize the trace, then read each credential off equation (1). The trace machinery (definition, linearity, basis-independence, cyclicity) is owned by the completeness article and consumed here by citation.

Step 1. Average the sandwich. For the ensemble, the mean of $\hat A$ is the classically weighted mean of the quantum means:

\langle\hat A\rangle = \sum_i p_i\,\langle\psi_i|\hat A|\psi_i\rangle = \sum_i p_i\,\mathrm{Tr}\bigl(|\psi_i\rangle\langle\psi_i|\hat A\bigr) = \mathrm{Tr}(\rho\,\hat A) \qquad (2)

Why the middle equality: $\langle\psi|\hat A|\psi\rangle = \mathrm{Tr}(|\psi\rangle\langle\psi|\hat A)$ — expand the trace in any basis and the sifting collapses it; linearity of the trace then pulls the sum inside. One formula now covers pure and mixed alike, and the Born rule rides along: $P(a_n) = \mathrm{Tr}(\rho\hat P_n)$ .

Step 2. Hermiticity: each projector is Hermitian and the $p_i$ are real, so $\rho^\dagger = \rho$ . Its eigenvalues are therefore real and its eigenvectors orthogonal (the Hermiticity theorems, recalled).

Step 3. Unit trace: $\mathrm{Tr}\rho = \sum_i p_i\,\mathrm{Tr}|\psi_i\rangle\langle\psi_i| = \sum_i p_i = 1$ — total probability, wearing trace clothing. (Check it against (2) with $\hat A = \mathbb 1$ .)

Step 4. Positivity: for any $|\phi\rangle$ , $\langle\phi|\rho|\phi\rangle = \sum_i p_i|\langle\phi|\psi_i\rangle|^2 \ge 0$ — every diagonal element in every basis is a legitimate probability. Why the converse matters more: any operator with these three credentials can be written as (1) — diagonalize it, and its spectral form IS an ensemble (eigenvectors with eigenvalue weights). The three properties are the complete definition of "state"; ensembles are one way to build them, not what they are. *Verified: numerically on a random 5-member ensemble in 4 dimensions — $\mathrm{Tr}(\rho\hat A)$ equals the weighted ensemble average to machine precision; Hermiticity, unit trace, positivity, and $\mathrm{Tr}\rho^2 < 1$ all hold.*

The purity $\mathrm{Tr}\rho^2$ grades the ignorance. For a pure state $\rho^2 = \rho$ (projectors are idempotent) so $\mathrm{Tr}\rho^2 = 1$ ; for anything genuinely mixed, $\mathrm{Tr}\rho^2 = \sum\lambda_k^2 < \sum\lambda_k = 1$ (eigenvalues strictly between 0 and 1 lose weight when squared), with the floor $1/d$ reached by the maximally mixed $\mathbb 1/d$ — total ignorance, all directions equal. Purity is basis-independent, computable in two matrix multiplications, and the first diagnostic run on any laboratory ρ.

The Bloch ball: the entire qubit state space, drawn

For a two-level system the credentials become a picture. Any $2\times2$ Hermitian unit-trace matrix expands uniquely as

\rho \;=\; \tfrac12\bigl(\mathbb 1 + \mathbf r\cdot\boldsymbol\sigma\bigr), \qquad \mathbf r = \bigl(\langle\sigma_x\rangle, \langle\sigma_y\rangle, \langle\sigma_z\rangle\bigr) \qquad (3)

where: the Bloch vector $\mathbf r$ is the state — its components are the three spin expectation values (take traces against each $\sigma_k$ to see it), so the abstract operator becomes an arrow you could measure component by component. Positivity is the geometry: the eigenvalues of (3) are $(1 \pm |\mathbf r|)/2$ (machine-verified), non-negative exactly when $|\mathbf r| \le 1$ — the state space is a solid ball. Purity is the radius: $\mathrm{Tr}\rho^2 = (1 + r^2)/2$ , so the surface $r = 1$ is the pure states (the kets you have used all track, one per direction), the interior is the mixtures, and the center $\mathbf r = 0$ is the unpolarized $\mathbb 1/2$ — zero spin expectation along every axis (checked on all three). Optics reads the same picture as the Poincaré sphere with $\mathbf r$ the Stokes vector; the worked example below computes one.

Figure

The Bloch ball — pure skin, mixed flesh, ignorant core

The ball also draws the non-uniqueness theorem. Mixing $|{+z}\rangle$ and $|{-z}\rangle$ fifty-fifty walks you along the vertical diameter to the center; mixing $|{+x}\rangle$ and $|{-x}\rangle$ walks the horizontal diameter — to the same center (machine-checked: both blends give exactly $\mathbb 1/2$ ). Infinitely many ensembles, one ρ; and since every prediction is $\mathrm{Tr}(\rho\hat A)$ , no experiment can tell the preparations apart — not now, not with future technology: the equality of operators is the end of the argument. The lesson cuts deep: a density operator is not a record of which ensemble happened; it is the complete catalog of what can be predicted, and distinct histories with one catalog are one state. (This very fact becomes a no-signalling guarantee in the entanglement chapter — remote steering can choose which ensemble realizes your ρ, but never change ρ itself.)

Forced mixtures: the partial-trace preview

Go deeper · Derivation B · 1 minHalf of a Bell pair is maximally mixed — the mixture nobody chose

Derivation BHalf of a Bell pair is maximally mixed — the mixture nobody chose

We compute the local state of one qubit from the entangled pure pair $|\Phi\rangle = (|{\uparrow\uparrow}\rangle + |{\downarrow\downarrow}\rangle)/\sqrt2$ , exhibiting the third and deepest factory of mixedness: entanglement. The tool is the partial trace — sum the composite ρ over an orthonormal basis of the unobserved partner — used here as a recipe on one example; its theory (why it is the unique local-description map) belongs to the entanglement chapter's reduced-density-matrix article, by name. Watch a pure whole produce a maximally mixed part.

Step 1. Build the whole: $\rho_{AB} = |\Phi\rangle\langle\Phi|$ — pure by construction, $\mathrm{Tr}\rho_{AB}^2 = 1$ (machine-checked). Expanded, it contains four blocks including the coherences $|{\uparrow\uparrow}\rangle\langle{\downarrow\downarrow}|$ between the branches.

Step 2. Trace out B: $\rho_A = \mathrm{Tr}_B\,\rho_{AB} = \sum_{s\,\in\,\{\uparrow\downarrow\}}\langle s|_B\,\rho_{AB}\,|s\rangle_B$ . The diagonal blocks each contribute half a projector; the coherence blocks need $\langle\uparrow|\downarrow\rangle_B = 0$ and die:

\rho_A = \tfrac12|{\uparrow}\rangle\langle{\uparrow}| + \tfrac12|{\downarrow}\rangle\langle{\downarrow}| = \tfrac{\mathbb 1}{2} \qquad (4)

Why the coherences vanished: they connect branches in which B is orthogonal — locally invisible, because certifying them requires touching both sides. The information is not lost; it is stored in correlations no local measurement reads.

Step 3. Read the result: qubit A alone is exactly the unpolarized state — center of the Bloch ball, purity $\tfrac12$ , fifty-fifty along every axis (all machine-verified). Identical to an oven's output, though the global state is pure and nobody is ignorant of anything.

Step 4. Bank the doctrine this exhibit seeds: mixedness of a part measures entanglement of the whole (quantified by entropy in the entanglement chapter); local measurements cannot distinguish "half a Bell pair" from "coin-flipped preparation" (the non-uniqueness theorem, now enforcing no-signalling); and decoherence — the environment entangling with everything — manufactures exactly this death-of-coherences in every unattended quantum system, which is why the off-diagonals of laboratory density matrices decay. The track's final chapter is this Step run in reverse. Verified: numerically — $\rho_{AB}$ pure ( $\mathrm{Tr}\rho^2 = 1$ ); partial trace equals $\mathbb 1/2$ exactly; local purity $= \tfrac12$ .

What ρ does not cover in this article, by ownership: its time evolution (the von Neumann equation $i\hbar\dot\rho = [\hat H, \rho]$ ) belongs to Properties and Time Evolution of the Density Matrix in the entanglement chapter, along with the full property audit at graduate register; entropy measures to the entanglement-entropy article; and the unread-measurement map $\rho' = \sum\hat P_n\rho\hat P_n$ was the collapse article's exhibit — all recalled by name, none re-derived.

Limiting cases

case	result	meaning
single-term ensemble	$\rho = \|\psi\rangle\langle\psi\|$ , $\mathrm{Tr}\rho^2 = 1$	pure states are the special case
orthogonal two-term 50/50 (qubit)	$\rho = \mathbb 1/2$	maximal ignorance; basis of mixing forgotten
non-orthogonal ensemble	still legal — diagonalize for the canonical form	ensembles overshoot states: many-to-one
$\mathrm{Tr}\rho^2$	from $1$ (pure) down to $1/d$	the ignorance dial
Bloch $\|\mathbf r\| = 1$ / $< 1$ / $> 1$	pure / mixed / illegal	positivity as geometry
part of an entangled pure state	mixed, via partial trace	mixedness without ignorance
classical limit of ρ	diagonal in one preferred basis	probability distribution recovered — decoherence's endpoint

Common trapA mixture is just a superposition whose phase I forgot to write down

The costliest conflation in the chapter, because the two objects agree on one measurement basis and disagree on the rest — so lazy checking "confirms" the error. The superposition $(|{+z}\rangle + |{-z}\rangle)/\sqrt2$ and the 50/50 mixture share their $z$ -statistics; rotate to $x$ and the pure state answers $+\hbar/2$ with certainty while the mixture stays at coin-flip ( $\langle S_x\rangle$ : $\hbar/2$ versus $0$ — machine-checked). Algebraically the receipts are the off-diagonal coherences: $\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$ versus $\tfrac12\begin{pmatrix}1&0\\0&1\end{pmatrix}$ . Two enforcement rules. First, averaging over an unknown phase produces the mixture, not a superposition: $\tfrac{1}{2\pi}\int d\varphi\,|\psi_\varphi\rangle\langle\psi_\varphi| = \mathbb 1/2$ — "random phase" is precisely phase forgotten, coherence dead. Second, the reverse error in computation: expanding a pure state's probabilities as if branches didn't interfere (dropping cross terms) silently replaces your state by its dephased mixture — the position-representation article's $P_{\rm left} = 0.92$ -not- $0.5$ example was exactly this trap with the cross term kept, correctly.

Worked examples

Example 1 — the interrogation table, with numbers (confidence builder). Compare the pure state $|{+x}\rangle$ against the 50/50 $z$ -mixture on every row of the standard audit.

Step 1. Build both operators: $\rho_{\rm pure} = |{+x}\rangle\langle{+x}| = \tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$ ; $\rho_{\rm mix} = \tfrac12|{+z}\rangle\langle{+z}| + \tfrac12|{-z}\rangle\langle{-z}| = \tfrac12\mathbb 1$ . Same diagonal; the difference is entirely coherence.

Step 2. $z$ -row: $\mathrm{Tr}(\rho\,\hat S_z) = 0$ for both, and both give 50/50 outcomes — indistinguishable, as advertised.

Step 3. $x$ -row: $\mathrm{Tr}(\rho_{\rm pure}\hat S_x) = \boxed{\hbar/2}$ (certainty) versus $\mathrm{Tr}(\rho_{\rm mix}\hat S_x) = \boxed{0}$ (coin flip) — one Stern–Gerlach rotation separates them completely (machine-verified).

Step 4. Purity row: $\mathrm{Tr}\rho_{\rm pure}^2 = 1$ , $\mathrm{Tr}\rho_{\rm mix}^2 = \tfrac12$ — the ignorance dial at its two ends. Bloch reading: $\mathbf r = \hat x$ (surface) versus $\mathbf r = 0$ (center).

Example 2 — a partially polarized photon beam (exam level). A beam is prepared as 70% vertically polarized photons and 30% polarized at $+45°$ . Find ρ, the purity, the Stokes (Bloch) vector and degree of polarization, and the probability of passing a vertical polarizer.

Step 1. Build: with $|V\rangle = \binom01$ , $|D\rangle = \tfrac{1}{\sqrt2}\binom11$ : $\rho = 0.7|V\rangle\langle V| + 0.3|D\rangle\langle D| = \begin{pmatrix}0.15&0.15\\0.15&0.85\end{pmatrix}$ (machine-checked). Note the ensemble's states are not orthogonal — perfectly legal.

Step 2. Purity: $\mathrm{Tr}\rho^2 = \boxed{0.79}$ — partially mixed, closer to pure than to $\tfrac12$ .

Step 3. Stokes vector: $r_x = 2\,\mathrm{Re}\,\rho_{01} = 0.3$ , $r_y = 0$ , $r_z = \rho_{00} - \rho_{11} = -0.7$ , so $|\mathbf r| = \sqrt{0.58} = \boxed{0.762}$ — the degree of polarization, consistent with purity via $(1 + r^2)/2 = 0.79$ ✓. Eigenvalues $(1 \pm 0.762)/2 = 0.881, 0.119$ : the beam is equivalent to an 88/12 mixture of two orthogonal elliptical polarizations — the canonical (diagonalized) ensemble, different from the preparation story, same ρ.

Step 4. Vertical-polarizer transmission: $P(V) = \langle V|\rho|V\rangle = \rho_{11} = \boxed{0.85}$ — the 70% that is V plus half of the 30% diagonal beam. Every optics bench measuring Stokes parameters is doing density-matrix tomography with lenses.

The teenager who invented ρ

Origins · 1 minLev Landau and how this was discovered

Lev Landau

1908–1968 · Leningrad, Kharkov, Moscow

This track met Landau once already — his one-dimensional theorems band lives at the non-degeneracy article, in Course-of-Theoretical-Physics costume. Here he is nineteen years old. In Das Dämpfungsproblem in der Wellenmechanik (1927), analyzing how a quantum system loses coherence to radiation it emits, Landau confronted exactly this article's problem: the subsystem of an entangled whole has no wavefunction of its own. His solution — describe it by the operator now called the density matrix, introduced in that paper for the first time anywhere — is the partial-trace insight, arrived at from the entanglement side rather than the ensemble side: forced mixtures before convenient ones. John von Neumann's independent treatment months later (and his 1932 Grundlagen) supplied the systematic theory — ensembles, entropy, measurement — which is why both names attach to the object; von Neumann's own bands in this chapter (spectral splitting, self-adjointness) are elsewhere, and the birth certificate belongs to the teenager from Baku. Landau's later career — superfluidity, phase transitions, the Nobel Prize of 1962, the school whose entrance exam ("theoretical minimum") terrorized generations — tends to eclipse the 1927 paper; this chapter, which has spent ten articles building toward exactly the object he scribbled down first, restores the billing.

History

1927

— Landau's damping paper introduces the density matrix for subsystems of entangled wholes

1927

— von Neumann, independently: ensembles,

\mathrm{Tr}(\rho\hat A)

, and the statistical operator

1932

— Mathematische Grundlagen: the three credentials, entropy

-\mathrm{Tr}\rho\ln\rho

, measurement as dephasing

1946

— Bloch's vector picture for two-level ensembles; magnetic resonance adopts the ball

1957

— Fano's review installs ρ as standard laboratory equipment (beams, polarization, resonance)

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §3.4 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Complement E_III · Nielsen & Chuang, Quantum Computation and Quantum Information, §2.4 · Blum, Density Matrix Theory and Applications, 3rd ed. · Landau, Z. Phys. 45, 430 (1927) · Fano, Rev. Mod. Phys. 29, 74 (1957).

Provenance. All equations are standard density-operator results. A verifier should recompute: (1) the three defining properties from the ensemble definition (Sakurai §3.4); (2) $\langle A\rangle = \mathrm{Tr}(\rho A)$ against the weighted ensemble average; (3) the purity criterion and the Bloch-ball positivity condition; (4) the partial trace of a Bell state. For this edition all four were machine-executed (numpy/sympy), plus this article's additions: the full interrogation table for $\mathbb 1/2$ versus $|{+x}\rangle$ ( $\langle S_x\rangle = 0$ vs $\hbar/2$ ; purities $\tfrac12$ vs 1; idempotence check); the ensemble non-uniqueness exhibit ( $z$ -blend $=$ $x$ -blend $= \mathbb 1/2$ exactly); Bloch eigenvalues $(1\pm|\mathbf r|)/2$ with positivity failing at $|\mathbf r| = 1.4$ ; the purity–radius law $(1+r^2)/2$ ; the polarization example ( $\rho = [[0.15, 0.15], [0.15, 0.85]]$ , purity $0.79$ , Stokes vector $(0.3, 0, -0.7)$ , $|\mathbf r| = 0.762$ , eigenvalues $0.881/0.119$ , $P(V) = 0.85$ ); the Bell partial trace ( $\rho_A = \mathbb 1/2$ exact, global purity 1, local purity $\tfrac12$ ); and the unpolarized state's zero expectation on all three axes. Per the chapter ownership map: trace machinery is owned by Completeness Relation and Resolution of Identity; outer-product projectors by Projection Operators; the unread-measurement dephasing map by Measurement and Wavefunction Collapse; time evolution (von Neumann equation), the graduate-register property audit, reduced-density-matrix theory, and entropy measures by the entanglement chapter's density-operator suite — The Density Operator and Mixed States (the register twin this UG serial anchors), Properties and Time Evolution of the Density Matrix, The Reduced Density Matrix and Partial Trace, and Entanglement Entropy and Measures of Entanglement, all recalled BY NAME.

Builds on: Hilbert Space and State Vectors · Linear Operators on Hilbert Space · Projection Operators · Completeness Relation and Resolution of Identity · Born Rule and Measurement Probabilities · The Postulates of Quantum Mechanics

Leads to: Measurement and Wavefunction Collapse · Quantum Entanglement · Entanglement Entropy and Measures of Entanglement

Is the density operator a state of the system or a state of my knowledge?

Operationally the question dissolves: ρ is the complete catalog of predictions, and two agents assigning the same ρ make identical bets regardless of their stories. The interesting content is in the non-uniqueness theorem: because many ensembles share one ρ, "which pure state the system is really in" is not experimentally meaningful for a mixed ρ — and for the entanglement-induced case there is no finer fact to be ignorant of (no member of any ensemble is "the real one" for half a Bell pair). Pragmatically: compute with ρ, and let the interpretations chapter at the track's end host the philosophy with proper referees.

Why is the mixture written with probabilities and projectors instead of adding the kets with random phases?

Because adding kets — with any phases — always produces another pure state, just a different point on the Bloch sphere's surface. Ignorance is not a superposition with a shrug; it is the average over the unknown parameter, and averaging the projector $|\psi_\varphi\rangle\langle\psi_\varphi|$ over a uniform random phase kills the cross terms exactly: $\tfrac{1}{2\pi}\int d\varphi\,(\cdots) = \tfrac12\mathbb 1$ for the equatorial family. The projector (density-matrix) language is forced precisely because probabilities attach to states, while amplitudes attach to branches within a state — the two-layer structure of the overview.

Can I always tell from ρ alone whether the mixedness came from an oven or from entanglement?

No — locally the two are perfectly indistinguishable, and that is a theorem, not a technological limitation: all predictions flow from ρ, and the ρ's are equal. The distinction is detectable globally: if a partner system exists, joint measurements reveal the correlations (Bell-inequality experiments are exactly this), while the oven's atoms have no partner to interrogate. The impossibility of telling locally is load-bearing physics — it is what prevents entanglement from transmitting signals, since a remote party's actions change which ensemble realizes your ρ but never ρ itself.

Do the p_i in the ensemble have to be the eigenvalues of ρ?

No — that is one ensemble among infinitely many, the canonical one obtained by diagonalizing (orthogonal members, eigenvalue weights, and the one Example 2's Step 3 extracts). Preparations are free to use non-orthogonal members with any weights (70% V plus 30% diagonal light, as in the example), and theorems exist classifying exactly which ensembles realize a given ρ (all mixtures related by an isometry on the "purification" — entanglement-chapter machinery). When a calculation needs a distinguished decomposition — entropy, for instance — it is always the eigen-decomposition.

How does ρ evolve in time, and why isn't it just the Schrödinger equation?

For a closed system, each member of the ensemble evolves unitarily, so $\rho(t) = U\rho(0)U^\dagger$ — differentiating gives the von Neumann equation $i\hbar\,\partial_t\rho = [\hat H, \rho]$ (note the commutator: ρ is an operator, so it evolves by conjugation, not by the ket equation directly; the sign is opposite to Heisenberg-picture observables). For open systems — the reason ρ exists — unitarity fails and the evolution acquires dissipative terms (Lindblad generators). Both stories are owned downstream: the closed case by the entanglement chapter's evolution article and the dynamics chapter's pictures, the open case beyond this track's core.

Is there a wavefunction-like object for mixed states, or is the matrix the only representation?

The matrix (operator) is fundamental, but useful re-dressings exist. Purification: every mixed ρ is the partial trace of some pure state on a doubled system — so you may always compute with a ket upstairs (the entanglement chapter proves and exploits this). Phase-space: the Wigner function represents ρ as a quasi-distribution over $(x, p)$ , beloved of quantum optics. And for qubits, the Bloch vector is a complete faithful coordinate. What does not exist is a single-system ket for a genuinely mixed state — that absence is the content of purity $< 1$ .

What is the entropy of a density matrix, and why is it not this article's business?

The von Neumann entropy $S = -\mathrm{Tr}(\rho\ln\rho) = -\sum_k\lambda_k\ln\lambda_k$ — zero for pure states, $\ln d$ for maximal mixing — upgrades the purity dial to a full information measure with operational meaning (compression, entanglement quantification, thermodynamics). It deserves and has its own article (Entanglement Entropy and Measures of Entanglement, final chapter) because its serious applications are entanglement bookkeeping: the entropy of a reduced state measures how entangled the part is with the rest. Here, purity suffices for every diagnostic this chapter needs.